Unsloth Releases 2-Bit GGUF Version of MiniMax-M3.1-Flash, Cutting Size to 143.9GB

Unsloth released a 2-bit GGUF quantized version of MiniMax-M3.1-Flash on October 27. The model, based on MiniMax's 428B-parameter MoE architecture, has its footprint reduced to 143.9GB, about 45% smaller than the original. In Unsloth's evaluation, the version retains 91.5% top-1 token agreement across seven domains.

The release aims to lower the barrier for self-hosting, supporting local deployment through Unsloth Desktop and llama.cpp, as well as an OpenAI-compatible API for existing applications and agents. With over 78,000 GitHub stars and more than 500 million model downloads on Hugging Face, Unsloth provides a path for MiniMax-M3.1-Flash into an established community of developers building with open models.